We propose an alternative bivariate zero-inflated negative binomial (BZINB) regression model based on a copula. The empirical result shows that the proposed model performs better than the existing BZINB models in terms of the maximum log-likelihood and the AIC.